05.21 Media Attribution in Practice: post-IDFA and Backend Event Definitions

Governance metadata | Owner: Sales Lead | Last reviewed: 2026-10-22 | Next review: 2027-01-31

Methodology is in 05.19; this document locks down the executable events, windows, dashboards, and red lines. Platform policies are governed by the official sources.

---

1. Default Four Definitions (One Shared Set of Definitions for the Whole Team)

Metric	Definition	Window	Use
Backend closed orders	Payment succeeded and not refunded by T+7	Order day	Real revenue
Attributed orders	Traceable via UTM / affiliate / dedicated code	7-day click / 1-day view	Channel contribution
Channel contribution margin	Attributed revenue − variable cost − channel fees − variable ad spend	Month	Whether to add budget
Unknown-attribution share	Untagged orders / total orders	Week	Data credibility

Red line (default): unknown attribution >40% → it is forbidden to cut brand/social budget on the basis of "last-click ROAS" alone.

---

2. post-IDFA Technology Stack (Conceptual)

1. Ad SDK + Conversion API to post back the payment (server-side first)
2. Site/App event tracking → server-side events
3. KOL dedicated code as a weak-attribution backup
4. Split out new vs returning customers, and brand vs non-brand terms
5. Sampling: manually reconcile 50 orders per week; if the error rate is ≥5%, fix the tracking first and only then talk about optimisation (align with §24)

---

3. Event Dictionary

Event	Trigger	Required Properties
page_view	Landing	utm_source/medium/campaign/content
add_to_cart	Add to cart	value, sku
begin_checkout	Begin checkout	value, items
purchase	Payment succeeded	order_id, revenue, currency, sku, channel_hint

Naming convention example: utm_source=kol_yt_nebula&utm_medium=affiliate&utm_campaign=astro_q4.

---

4. Dashboard Fields (Interface with 01.13 / KPI Base)

Block	Fields
Spend	Daily spend, platform, campaign
Platform efficiency	ROAS, CPA, CTR, ACOS (Amazon single track)
Backend efficiency	Backend ROAS, post-refund ROAS
Credibility	Unknown-attribution %, dedicated-code order %, server-side post-back success rate
Decision	Whether the circuit-breaker / budget-increase conditions are met

Forbidden: adding Meta ROAS and Amazon ACOS together and treating the result as the total efficiency.

---

5. Scenario Playbooks

Crowdfunding Warm-Up (05.02.01)

• VIP code contribution share, CPV
• Free leads serve only the upper funnel and do not count toward "verified paid"

Major Promo (05.05 / 05.18)

• 72h around the lock-in window: backend ROAS, unknown-attribution %
• Recompute the contribution margin whenever the Deal price changes

KOL (04.03)

• Seeding ROI = attributed orders or dedicated code; it is forbidden to look at view counts alone

---

7. Related

05.19 / 05.18 / 01.13 / 05.04 / 05.02

---

7. Attribution Dispute Adjudication Order (Default)

1. Total backend closed revenue (finance)
2. Traceable order-level detail
3. Server-side event success rate
4. Platform reports (for optimisation reference only)
5. When unknown attribution is >40%, budget decisions are downgraded to "test budget"

8. Division of Labour with BI (05.19)

• 05.19: models, windows, multi-touch theory
• This document: events, tracking, dashboards, red lines
• When implementation conflicts: backend payment facts prevail

9. Common Tracking Failures

Failure	Symptom	Fix
Front-end only	iOS orders lost	Server-side post-back
Duplicate purchase	Inflated ROAS	Deduplicate order_id
Timezone error	Day-boundary misalignment	Store everything in UTC
Code hijacked	Channels steal each other's orders	Code rules + risk control

---

8. Server-Side Post-Back Acceptance Steps

1. Make 1 sandbox payment → check the event table
2. Whether the ad backend "Event Management" received it
3. Whether duplicate payments are deduplicated
4. Spot-check the timezone and currency fields

9. Budget Decision Matrix (Default)

Unknown Attribution	Backend ROAS	Action
<20%	On target	Budget can be increased
20–40%	On target	Increase in small steps + intensify attribution
>40%	Any	Fix tracking/codes first; freeze any brand-spend cuts

10. With 04 Seeding

The primary column of the seeding ROI table: dedicated-code orders; the secondary column: the short UTM window; view counts are a process metric only.

---

11. Full Channel Naming Convention Table

Channel	utm_source Example	Notes
Meta	meta
Google	google
YouTube KOL	yt_	+ dedicated code
On-site	onsite	brand terms
Email	email_
Offline	offline_	QR code

12. Post-Refund ROAS

Post-refund ROAS = (closed orders − T+7 refunds) / spend

A version must be produced within 14 days after a major promo, to prevent "inflated same-day numbers".

13. Differences from the 05.05 Major Promo

	Regular	Major Promo
Attribution window	7-day click	Can be shortened to 1–3 days with a note
Unknown attribution	Monitored	Allowed to rise, but decisions are downgraded
Budget	By ROAS	Dual objective of GMV + contribution margin

14. Data Quality Score (Weekly)

Item	Score
Server-side post-back success	30
Code/UTM coverage	30
Unknown attribution <40%	20
Both ROAS views present	20

<80: fix the data first, then talk about large-scale media spend.

15. Tracking Change Management

• Any event field change goes through a PR + a regression test case
• Change week: the dashboard is marked "data change"
• Operations staff are forbidden to rename events on their own

16. Field Alignment with the 04.03 Seeding Table

Add these columns to the seeding table: Dedicated Code, UTM content, Attributed Orders, Backend Revenue.

18. Platform Comparison Table (Conceptual)

Platform	Attribution Characteristics	Caveats
Meta	The Conversion API matters	iOS losses
Google	Search + shopping	Split out brand terms
Amazon on-platform	ACOS single track	Do not mix-add with Meta
TikTok	Content-driven	Code + UTM

19. Test Design Principles

• Single-variable weekly tests
• Budget no smaller than the statistically meaningful threshold (set by the business)
• During mixed-attribution periods, do not "fully pause brand terms"

20. With 07.10 AI

If the landing page uses AI-generated creative: the conversion event still uses purchase as the source of truth, and the generation cost is booked to marketing or R&D (decided by finance).

21. Crisis: Data Platform Outage

Post-back outage >24h: pause the automated rules and switch to manual daily reports; backfill the events after recovery.

23. Event Post-Back Acceptance Sheet (Before Launch)

Check	Pass Standard	Action on Failure
1 sandbox payment	A purchase exists in the event table	Scaling up is forbidden
order_id unique	No duplicates	Fix the deduplication
Amount = actual paid	0 error	Fix the field
Server-side post-back	Visible in the ad backend	Fix the API
Timezone	Consistent with the finance day boundary	Store everything in UTC
50-order sampling reconciliation	Error rate <5%	Fix the UTM rules

24. Budget-Increase Gate (Default, All Conditions Must Be Met)

1. Backend ROAS ≥ target (filled in by the project)
2. Unknown attribution <40%
3. Server-side post-back success rate >95%
4. Post-refund ROAS still ≥ target×0.8
5. No ACOS circuit breaker triggered (if Amazon)

If any one condition is not met → only a probe budget of ≤10% is allowed.

25. Budget-Cut Gate (Default)

Situation	Action
Unknown attribution >40% with no remediation plan	Forbidden to cut brand/content; cut only paid search with poor attribution
Single-channel ROAS poor for 3 days	Reduce by 50%, do not fully pause
Data quality score <80	Fix the data first; freeze structural budget cuts

26. Field Mapping to the 05.19 Model

Field in This Document	05.19
Attributed orders	The "trackable conversion" in multi-touch
Unknown attribution	The residual outside the model
Post-refund ROAS	The finance-aligned column

In case of conflict: payment facts > platform reports > model inferences.

27. Checklist (Final)

• [ ] All six items on the acceptance sheet have passed

• [ ] The budget-increase / budget-cut gates have been posted to the media team channel

• [ ] The 05.19 field-mapping table has been trained